Explore More Automation

Markus Guerster reminds readers at the beginning of his book, "Artificial Intelligence Will Revolutionize Manufacturing," that Industry 4.0 is still in the early stages. About a decade ago, companies began to adopt digital technologies, including networking machines and processes, to optimize plant production. The byproduct of Industry 4.0 is significant amounts of enterprise data that companies didn't have to manage in the past.

Fast forward to 2024, big data and varying degrees of data maturity are present, and now food manufacturers want to use Industry 4.0 to develop artificial intelligence (AI) strategies. However, it's happening fast, similar to Industry 4.0 claims from 10 years ago.

Guerster's book, released in 2024, offers a step-by-step look at the leadup to AI technology while educating readers on machine learning, deep learning, generative AI and ChatGPT technologies. Guerster also touches on the Industry 4.0 flywheel, return on investment (ROI), the connection between machine learning and AI, neural networks and unsupervised learning.

Below is our wide-ranging discussion on his book, food manufacturers' understanding of AI-techologies and what's in store for 2025.

FOOD ENGINEERING: Industry 4.0 started with significant investments in networking, sensor and connected equipment. It's happening again with AI. Where is the food industry with AI adoption going into 2025?

Markus Guerster: I poked into a little bit of a dark hole (with the book) and I would say there's almost community peer pressure to do something with AI manufacturing. On the other hand, there's a lack of good content on how to get started, and this is how I positioned the book.

Over the last few months, the C-Suite's attitude has shifted on AI investments. At first, it was exciting and let's try it; now here's another AI product. We're beyond the hype cycle phase – many companies tried it, and it didn't quite work, so we're in an interesting phase now. Regarding technology adoption, this AI phase is standard: There's something new, everybody jumps in, and the marketing team exploits it. People get burned slowly and then start understanding what works and what doesn't.

FE: Similar to Industry 4.0 pilot projects, executive teams are leaning toward agile approaches for AI technology as learning continues and expectations mature. What are you seeing with food manufacturers?

MG: I firmly believe in agile methods – start with one problem, try to make it as small as possible while solving a pain point and then work on a solution. Then, keep working until it solves the problem, keep iterating on it, and find the next one. The mindset slowly shifts in that direction when plant managers see a small thing can be piloted.

It's also becoming a lot cheaper to pilot AI-based technologies. In the past couple of years, if you wanted to try, some AI companies had to hire big internal teams or a prominent cloud provider. But, these days, you can just try it. Very, very simple. ChatGPT team is a classic example where it's basically free to try AI, even though it doesn't really help the plant management to run the operations faster. However, minimal resources allow you to get small things off the ground a lot faster, and you keep on learning.

FE: As mentioned in the book, machine learning is the heartbeat of modern AI. So how do companies approach data governance and maturity so they can prepare for the stern requirements of AI modeling?

MG: Good data is the absolute foundation of everything that you're building on top of this (AI). If it's AI, machine learning, or even simpler statistics or visualization, bad data means poor decisions – trash in, trash out. When we started MontBlancAI two and a half years ago, we had this naive view that manufacturers had good data sources since machines are excellent data sources. So, manufacturers must have their data cleaning and aggregation storage figured out, and we can add the AI machine learning layer on top of this (machine layer). We learned that the data sources are great, but the technology that collects, cleans and stores the machine information is not good at handling significant data volume.

We developed a solution around our core machine-learning AI model that goes into data collection from the PLCs and sensors. This technology has been a way to get those small projects started.

Larger companies have a budget for an internal and big data team. They have to understand the whole data infrastructure, data governance and everything around this, and then go into use cases. But by the time this is done, two years, millions of dollars and no value has been generated. That's not the right way to do it. The solution should be vertical but very thin on the data side.

FE: What is happening with ROI in large and medium-sized food manufacturers regarding data governance and AI investments?

MG: For larger companies, a data foundation is very crucial. This understanding is also happening with small- to medium-sized food manufacturers, and that's important. However, larger food companies are feeling pressure around data governance ROI. Initially, the board was very patient and said let's invest two to three years and a couple of million in building a sound data foundation since it's essential for the future. Now, boards are putting the screws on and are asking about these significant investments. What are going to do now with the data that we have it? What do we do with it?

Smaller companies are seeing this and learning from this “clean and organize the data first” approach. These companies need to be quicker in getting an ROI calculated. It is a natural step to first clean and organize the data, spend a couple of months doing it and then a use case will develop. Flipping it around makes much more sense because you know which data you need to collect. That's a general trend, and these use cases provide much value to our customers in the food industry because they have processes and machines running.

Food processors want to understand whether this process is good or bad today and whether the machine is running good or bad today. With machine learning, AI learns this is how your process and machine behaves when it's going well. If there's any deviation from this, AI will put a red light up and notify you before something goes wrong.